--- title: "PowerPoint and Google Slides MCP Server for Reports | Rollstack" description: "Connect Claude, ChatGPT or any MCP client to Rollstack. Build, refresh and send PowerPoint and Google Slides reports from any data source." canonical: https://www.rollstack.com/mcp --- # Build, refresh and send trusted reports at scale from the AI tool you already use Connect Claude, ChatGPT or any MCP client to Rollstack. Build, edit, refresh and send PowerPoint and Google Slides reports from any data source, every number traced to its source, one version per client, region or team, right in the chat. MCP server URL: `https://api.rollstack.com/public/v1/mcp` (Streamable HTTP, OAuth sign-in) ## Start with one prompt ``` Help me connect the Rollstack MCP server at https://api.rollstack.com/public/v1/mcp if it isn't connected yet. Once it's connected, read docs.rollstack.com/api/mcp-server and suggest a first report I could build from my data. ``` ## What you can do with it Build slide decks from your company's data in your own template, refresh them on a schedule, and email each client, region or team its own version. - Build a QBR, business review, board pack or any deck from your data, in your template - Bursts: one version per client, region or team from one template, at any scale - Change a deck: add a slide, compare regions, rewrite titles, switch templates - Refresh a report's data from its sources - Schedule it on any cadence, hourly to yearly - Email it to its recipients, or send a test first - Export to PowerPoint, Google Slides or PDF - Ask why an update failed, or who received it ## Any data source Rollstack connects to any data source. The ones below are set up in a click; anything else connects through the API. | Kind | Sources | | --- | --- | | Data warehouses and lakehouses | Snowflake, Databricks, Google BigQuery, Amazon Redshift, Microsoft Fabric, Azure Synapse Analytics, Oracle Autonomous Data Warehouse, Teradata, ClickHouse, DuckDB, and more | | Semantic layers | dbt Semantic Layer, LookML (Looker), Snowflake semantic views, Databricks metric views, Cube, AtScale, and more | | BI tools | Microsoft Power BI, Tableau, Qlik Sense, Looker, ThoughtSpot, Domo, Sigma, Metabase, Amazon QuickSight, Mode, Omni, Hex, Lightdash, and more | | CRMs and business applications | Salesforce, Microsoft Dynamics 365, HubSpot, SAP, Oracle NetSuite, Workday, Zendesk, Gainsight, Google Analytics, and any other application through its API | | Spreadsheets and files | Excel, Google Sheets, CSV, PDFs, Word and Google Docs, files in SharePoint, OneDrive and Google Drive, JSON, and more | Connections are set up once in Rollstack, with your own credentials, and every agent you connect builds on them with your access. ## Why use Rollstack with your AI assistant A general assistant, even one connected to your warehouse or semantic layer, generates the deck from scratch every time: it writes the query, computes the numbers in the chat and lays out the slides, so the result changes from run to run and someone re-checks every figure before it goes out. Rollstack is what the assistant calls when the numbers have to be accurate, traceable, and ones you can explain and stand behind. ### Trusted by construction Every number comes from a deterministic query on the source and links back to it. It is pulled from the governed object your team already trusts, the dashboard view, the metric or the table, and lands in the slide unchanged, with its lineage: the source, the query and the last refresh. Anyone can trace a figure and verify it, so the deck shows the same numbers as the dashboard. AI builds the slides and drafts the commentary; it never computes a number. ### Refresh, don't regenerate Refreshes rerun the same queries, with no AI regeneration: the same deterministic logic every run, so every number is explainable and none drifts. The deck is mapped to its sources once, in your own PowerPoint or Google Slides template, on brand, with native, editable charts; each refresh updates the numbers in place, and your layout, edits and commentary stay where you left them. ### Scale and delivery, built in One template becomes a version per client, region or team (a burst), each with its own data, on any schedule. Delivery is built in: Rollstack emails each version to its recipients, with nothing to configure outside it. Recipients get a finished file: no license, no login, no access to your BI. ### Governed by default Access follows your permissions in Rollstack and in the connected sources, so each version of a deck holds only what its audience may see. Every run and every send is logged, and every version sent is archived. SOC 2 Type II, HIPAA and GDPR. Built for the decks where every number gets checked: client QBRs and business reviews, recurring client performance reports, weekly and monthly business reviews (WBR, MBR), leadership and department performance decks, KPI and operational status decks, marketing performance reports, board and investor decks, budget and forecast decks, financial flash and actuals reports, and governance, risk and compliance packs. | | Your assistant on its own | With Rollstack | | --- | --- | --- | | Where do the numbers come from? | Generated by the model | A deterministic query on your warehouse, BI tool or any connected source | | Can every number be traced to its source? | You'd check every number | Yes: each keeps its lineage (source, query, last refresh), so anyone can verify it | | Same logic on every refresh? | Generated again each time | Yes: the same deterministic logic each run, so every number is explainable and none drifts | | What happens next quarter? | You'd rebuild it | It refreshes with the logic you approved, on schedule, with no prompt and no re-check | | Is it on brand? | Its default style, or a template you re-apply each time | Your own PowerPoint or Google Slides template, with native, editable charts | | How does it reach people? | You download it and send it yourself | Emailed to each recipient from Rollstack, on schedule, with nothing to set up | | Who sees what? | Whatever the chat was given | Your permissions in Rollstack and in the connected sources; every run and every send logged | | One deck, or one per account? | One deck at a time | A burst: one version per client, region or team, at any scale | ## Works within your permissions - **Hosted endpoint.** Nothing to install or run. Add one URL to your AI tool. - **Sign in with OAuth.** Each sign-in lasts 90 days, and you can revoke it anytime. - **Your role, your approvals.** It does only what you approve, within your role. Deleting reports and managing members or connections stay in the app. - **Included on every plan.** Every Rollstack plan has it, with per-user limits. OAuth 2.1 with scopes you choose at sign-in; revoke any time under Settings → Connected apps. Any change the slides agent proposes, such as a schedule or an email send, waits for your approval. [SOC 2 Type II, HIPAA and GDPR.](https://www.rollstack.com/security) Data stays in your sources; Rollstack stores the decks. ## Steps for each tool ### Claude Code - Run: ``` claude mcp add --transport http rollstack https://api.rollstack.com/public/v1/mcp ``` - Run `/mcp` and sign in. ### Claude - Add a custom connector with the server URL. - Sign in to Rollstack and approve the access. On Team and Enterprise, an owner adds it first. ### ChatGPT - In settings, add a custom MCP server with the server URL (Streamable HTTP). - Sign in to Rollstack and approve the access. On Business and Enterprise, an admin allows custom servers first. ### Cursor [Add to Cursor](https://cursor.com/en-US/install-mcp?name=Rollstack&config=eyJ1cmwiOiJodHRwczovL2FwaS5yb2xsc3RhY2suY29tL3B1YmxpYy92MS9tY3AifQ==) - Then sign in when Cursor asks. ### VS Code [Add to VS Code](vscode:mcp/install?%7B%22name%22%3A%22rollstack%22%2C%22type%22%3A%22http%22%2C%22url%22%3A%22https%3A%2F%2Fapi.rollstack.com%2Fpublic%2Fv1%2Fmcp%22%7D) On GitHub Copilot Business and Enterprise, an admin turns on MCP servers first. ### Hermes Agent - Add it to `~/.hermes/config.yaml`: ``` mcp_servers: rollstack: url: "https://api.rollstack.com/public/v1/mcp" auth: oauth ``` - Run `hermes mcp login rollstack` and sign in. ### OpenClaw - Add it under `mcp.servers` in your OpenClaw config: ``` rollstack: { url: "https://api.rollstack.com/public/v1/mcp", transport: "streamable-http", auth: "oauth" } ``` - Stop the gateway, run `openclaw mcp login rollstack` and sign in, then start it again. ### Other clients - Create an API key under Settings → Connected apps. - Send it as a bearer token. ## Questions ### What is the Rollstack PowerPoint MCP server, and what can it do? MCP, the Model Context Protocol, is an open standard that lets AI assistants use tools in other apps. The Rollstack MCP server gives your assistant tools for reports, datasets, sharing, schedules, email and Bursts. To teach your agent common Rollstack workflows, add the [open-source skills](https://github.com/Rollstack/skills): `npx skills add Rollstack/skills`. ### Can Claude create PowerPoint presentations from my Tableau or Power BI dashboards? Yes. Rollstack is the AI presentation platform for business reporting. Connect Claude to it, and Rollstack builds the deck from any source connected in Rollstack: your warehouse, semantic layer, BI tool, CRM, spreadsheets or files. Claude can then have it refreshed, scheduled on any cadence, burst into one version per client, region or team, sent, and exported to PowerPoint or Google Slides. ### How is this different from Claude for PowerPoint? Claude for PowerPoint is Anthropic's add-in for working on a deck open in PowerPoint. The Rollstack MCP server connects Claude to your reporting: your connected data, your templates, and the schedules and emails that keep each report current. ### Why use Rollstack with Claude or ChatGPT? Claude and ChatGPT are good for drafting a one-off deck, but the figures are generated, so someone checks every number and rebuilds the deck next quarter. Rollstack is best when the numbers have to be trusted: every chart and figure comes from a deterministic query on any connected source (your warehouse, semantic layer, BI tool, CRM, spreadsheets or files), not from the model, and links back to it, so anyone can trace and verify it. Rollstack's agents build the deck in your template, refresh it on any schedule without regenerating it, and burst it into one version per client, region or team, at any scale. That is why teams use it for client QBRs, weekly and monthly business reviews, board and investor decks, KPI and performance decks, and budget, forecast and financial reports built from their data. ### Is it included on my plan, and are there limits? A Rollstack account in a Rollstack organization is required; the MCP server is included on every plan. Per-user limits include 6 messages a minute to the slides agent and 10 exports and slide-image calls a minute, across all your AI tools. ### Can my admin control it? In Rollstack, your assistant gets only the access you approve, and members can't grant access that only owners and admins hold. In your AI tool, owners on Claude Team and Enterprise, and admins on ChatGPT Business and Enterprise and GitHub Copilot Business and Enterprise, decide whether it can be added. ### Which data sources work? Any data source. Warehouses (Snowflake, Databricks, BigQuery, Redshift, Microsoft Fabric, Azure Synapse, Oracle, Teradata, ClickHouse, DuckDB and more), semantic layers (dbt, LookML, Snowflake semantic views, Databricks metric views, Cube, AtScale and more), BI tools (Power BI, Tableau, Qlik, Looker, ThoughtSpot, Domo, Sigma, Metabase, QuickSight, Mode, Omni, Hex, Lightdash and more), CRMs and business applications (Salesforce, Dynamics 365, HubSpot, SAP, NetSuite, Workday, Zendesk, Gainsight, Google Analytics and more), spreadsheets and files (Excel, Google Sheets, CSV, PDFs, Word, Google Docs, SharePoint, OneDrive, Google Drive, JSON and more) connect in a click; anything else, including internal systems, connects through the API. ### Can my own agent or pipeline call it? Yes. Any MCP client connects to https://api.rollstack.com/public/v1/mcp with OAuth 2.1, and scripts use the REST API with an API key; the OpenAPI spec and the auth guide at rollstack.com/auth.md cover both. The same tools build, refresh, schedule, email and export reports, and burst them into one version per client, region or team, at any scale. ## Go deeper - [Connect an AI assistant](https://docs.rollstack.com/api/connect-an-ai-assistant): Setup for each AI tool, step by step. - [MCP server reference](https://docs.rollstack.com/api/mcp-server): Every tool your assistant can call. - [Scopes and permissions](https://docs.rollstack.com/api/scopes): What each scope lets your assistant do. ### For developers - MCP server: `https://api.rollstack.com/public/v1/mcp` (Streamable HTTP). `initialize` and `tools/list` work without sign-in, so you can review the tools before connecting. - [MCP server reference](https://docs.rollstack.com/api/mcp-server) · [OpenAPI spec](https://api.rollstack.com/public/v1/openapi.json) · [How agents get credentials](https://www.rollstack.com/auth.md) - [Agent skills](https://github.com/Rollstack/skills) (MIT): `npx skills add Rollstack/skills` - MCP Registry: `com.rollstack/mcp` · server card: `/.well-known/mcp/server-card.json`